play kube: service container

Add the notion of a "service container" to play kube.  A service
container is started before the pods in play kube and is (reverse)
linked to them.  The service container is stopped/removed *after*
all pods it is associated with are stopped/removed.

In other words, a service container tracks the entire life cycle
of a service started via `podman play kube`.  This is required to
enable `play kube` in a systemd unit file.

The service container is only used when the `--service-container`
flag is set on the CLI.  This flag has been marked as hidden as it
is not meant to be used outside the context of `play kube`.  It is
further not supported on the remote client.

The wiring with systemd will be done in a later commit.

Signed-off-by: Valentin Rothberg <vrothberg@redhat.com>

5 files changed, 182 insertions, 98 deletions

diff --git a/pkg/domain/entities/play.go b/pkg/domain/entities/play.go
index bf7c33f2b..f1ba21650 100644
--- a/pkg/domain/entities/play.go
+++ b/pkg/domain/entities/play.go
@@ -54,6 +54,8 @@ type PlayKubeOptions struct {
 	LogOptions []string
 	// Start - don't start the pod if false
 	Start types.OptionalBool
+	// ServiceContainer - creates a service container that is started before and is stopped after all pods.
+	ServiceContainer bool
 	// Userns - define the user namespace to use.
 	Userns string
 }
diff --git a/pkg/domain/infra/abi/play.go b/pkg/domain/infra/abi/play.go
index 019361694..420d51483 100644
--- a/pkg/domain/infra/abi/play.go
+++ b/pkg/domain/infra/abi/play.go
@@ -28,12 +28,54 @@ import (
 	"github.com/containers/podman/v4/pkg/specgenutil"
 	"github.com/containers/podman/v4/pkg/util"
 	"github.com/ghodss/yaml"
+	"github.com/opencontainers/go-digest"
 	"github.com/pkg/errors"
 	"github.com/sirupsen/logrus"
 	yamlv2 "gopkg.in/yaml.v2"
 )
 
-func (ic *ContainerEngine) PlayKube(ctx context.Context, body io.Reader, options entities.PlayKubeOptions) (*entities.PlayKubeReport, error) {
+// createServiceContainer creates a container that can later on
+// be associated with the pods of a K8s yaml.  It will be started along with
+// the first pod.
+func (ic *ContainerEngine) createServiceContainer(ctx context.Context, name string) (*libpod.Container, error) {
+	// Similar to infra containers, a service container is using the pause image.
+	image, err := generate.PullOrBuildInfraImage(ic.Libpod, "")
+	if err != nil {
+		return nil, fmt.Errorf("image for service container: %w", err)
+	}
+
+	ctrOpts := entities.ContainerCreateOptions{
+		// Inherited from infra containers
+		ImageVolume:      "bind",
+		IsInfra:          false,
+		MemorySwappiness: -1,
+		// No need to spin up slirp etc.
+		Net: &entities.NetOptions{Network: specgen.Namespace{NSMode: specgen.NoNetwork}},
+	}
+
+	// Create and fill out the runtime spec.
+	s := specgen.NewSpecGenerator(image, false)
+	if err := specgenutil.FillOutSpecGen(s, &ctrOpts, []string{}); err != nil {
+		return nil, fmt.Errorf("completing spec for service container: %w", err)
+	}
+	s.Name = name
+
+	runtimeSpec, spec, opts, err := generate.MakeContainer(ctx, ic.Libpod, s, false, nil)
+	if err != nil {
+		return nil, fmt.Errorf("creating runtime spec for service container: %w", err)
+	}
+	opts = append(opts, libpod.WithIsService())
+
+	// Create a new libpod container based on the spec.
+	ctr, err := ic.Libpod.NewContainer(ctx, runtimeSpec, spec, false, opts...)
+	if err != nil {
+		return nil, fmt.Errorf("creating service container: %w", err)
+	}
+
+	return ctr, nil
+}
+
+func (ic *ContainerEngine) PlayKube(ctx context.Context, body io.Reader, options entities.PlayKubeOptions) (_ *entities.PlayKubeReport, finalErr error) {
 	report := &entities.PlayKubeReport{}
 	validKinds := 0
 
@@ -67,6 +109,30 @@ func (ic *ContainerEngine) PlayKube(ctx context.Context, body io.Reader, options
 			return nil, errors.Wrap(err, "unable to read kube YAML")
 		}
 
+		// TODO: create constants for the various "kinds" of yaml files.
+		var serviceContainer *libpod.Container
+		if options.ServiceContainer && (kind == "Pod" || kind == "Deployment") {
+			// The name of the service container is the first 12
+			// characters of the yaml file's hash followed by the
+			// '-service' suffix to guarantee a predictable and
+			// discoverable name.
+			hash := digest.FromBytes(content).Encoded()
+			ctr, err := ic.createServiceContainer(ctx, hash[0:12]+"-service")
+			if err != nil {
+				return nil, err
+			}
+			serviceContainer = ctr
+			// Make sure to remove the container in case something goes wrong below.
+			defer func() {
+				if finalErr == nil {
+					return
+				}
+				if err := ic.Libpod.RemoveContainer(ctx, ctr, true, false, nil); err != nil {
+					logrus.Errorf("Cleaning up service container after failure: %v", err)
+				}
+			}()
+		}
+
 		switch kind {
 		case "Pod":
 			var podYAML v1.Pod
@@ -90,7 +156,7 @@ func (ic *ContainerEngine) PlayKube(ctx context.Context, body io.Reader, options
 				podYAML.Annotations[name] = val
 			}
 
-			r, err := ic.playKubePod(ctx, podTemplateSpec.ObjectMeta.Name, &podTemplateSpec, options, &ipIndex, podYAML.Annotations, configMaps)
+			r, err := ic.playKubePod(ctx, podTemplateSpec.ObjectMeta.Name, &podTemplateSpec, options, &ipIndex, podYAML.Annotations, configMaps, serviceContainer)
 			if err != nil {
 				return nil, err
 			}
@@ -104,7 +170,7 @@ func (ic *ContainerEngine) PlayKube(ctx context.Context, body io.Reader, options
 				return nil, errors.Wrap(err, "unable to read YAML as Kube Deployment")
 			}
 
-			r, err := ic.playKubeDeployment(ctx, &deploymentYAML, options, &ipIndex, configMaps)
+			r, err := ic.playKubeDeployment(ctx, &deploymentYAML, options, &ipIndex, configMaps, serviceContainer)
 			if err != nil {
 				return nil, err
 			}
@@ -148,7 +214,7 @@ func (ic *ContainerEngine) PlayKube(ctx context.Context, body io.Reader, options
 	return report, nil
 }
 
-func (ic *ContainerEngine) playKubeDeployment(ctx context.Context, deploymentYAML *v1apps.Deployment, options entities.PlayKubeOptions, ipIndex *int, configMaps []v1.ConfigMap) (*entities.PlayKubeReport, error) {
+func (ic *ContainerEngine) playKubeDeployment(ctx context.Context, deploymentYAML *v1apps.Deployment, options entities.PlayKubeOptions, ipIndex *int, configMaps []v1.ConfigMap, serviceContainer *libpod.Container) (*entities.PlayKubeReport, error) {
 	var (
 		deploymentName string
 		podSpec        v1.PodTemplateSpec
@@ -170,7 +236,7 @@ func (ic *ContainerEngine) playKubeDeployment(ctx context.Context, deploymentYAM
 	// create "replicas" number of pods
 	for i = 0; i < numReplicas; i++ {
 		podName := fmt.Sprintf("%s-pod-%d", deploymentName, i)
-		podReport, err := ic.playKubePod(ctx, podName, &podSpec, options, ipIndex, deploymentYAML.Annotations, configMaps)
+		podReport, err := ic.playKubePod(ctx, podName, &podSpec, options, ipIndex, deploymentYAML.Annotations, configMaps, serviceContainer)
 		if err != nil {
 			return nil, errors.Wrapf(err, "error encountered while bringing up pod %s", podName)
 		}
@@ -179,7 +245,7 @@ func (ic *ContainerEngine) playKubeDeployment(ctx context.Context, deploymentYAM
 	return &report, nil
 }
 
-func (ic *ContainerEngine) playKubePod(ctx context.Context, podName string, podYAML *v1.PodTemplateSpec, options entities.PlayKubeOptions, ipIndex *int, annotations map[string]string, configMaps []v1.ConfigMap) (*entities.PlayKubeReport, error) {
+func (ic *ContainerEngine) playKubePod(ctx context.Context, podName string, podYAML *v1.PodTemplateSpec, options entities.PlayKubeOptions, ipIndex *int, annotations map[string]string, configMaps []v1.ConfigMap, serviceContainer *libpod.Container) (*entities.PlayKubeReport, error) {
 	var (
 		writer      io.Writer
 		playKubePod entities.PlayKubePod
@@ -374,6 +440,10 @@ func (ic *ContainerEngine) playKubePod(ctx context.Context, podName string, podY
 		}
 	}
 
+	if serviceContainer != nil {
+		podSpec.PodSpecGen.ServiceContainerID = serviceContainer.ID()
+	}
+
 	// Create the Pod
 	pod, err := generate.MakePod(&podSpec, ic.Libpod)
 	if err != nil {
diff --git a/pkg/specgen/generate/pause_image.go b/pkg/specgen/generate/pause_image.go
new file mode 100644
index 000000000..4aba230a3
--- /dev/null
+++ b/pkg/specgen/generate/pause_image.go
@@ -0,0 +1,89 @@
+package generate
+
+import (
+	"context"
+	"fmt"
+	"io/ioutil"
+	"os"
+
+	buildahDefine "github.com/containers/buildah/define"
+	"github.com/containers/common/pkg/config"
+	"github.com/containers/podman/v4/libpod"
+	"github.com/containers/podman/v4/libpod/define"
+)
+
+// PullOrBuildInfraImage pulls down the specified image or the one set in
+// containers.conf.  If none is set, it builds a local pause image.
+func PullOrBuildInfraImage(rt *libpod.Runtime, imageName string) (string, error) {
+	rtConfig, err := rt.GetConfigNoCopy()
+	if err != nil {
+		return "", err
+	}
+
+	if imageName == "" {
+		imageName = rtConfig.Engine.InfraImage
+	}
+
+	if imageName != "" {
+		_, err := rt.LibimageRuntime().Pull(context.Background(), imageName, config.PullPolicyMissing, nil)
+		if err != nil {
+			return "", err
+		}
+		return imageName, nil
+	}
+
+	name, err := buildPauseImage(rt, rtConfig)
+	if err != nil {
+		return "", fmt.Errorf("building local pause image: %w", err)
+	}
+	return name, nil
+}
+
+func buildPauseImage(rt *libpod.Runtime, rtConfig *config.Config) (string, error) {
+	version, err := define.GetVersion()
+	if err != nil {
+		return "", err
+	}
+	imageName := fmt.Sprintf("localhost/podman-pause:%s-%d", version.Version, version.Built)
+
+	// First check if the image has already been built.
+	if _, _, err := rt.LibimageRuntime().LookupImage(imageName, nil); err == nil {
+		return imageName, nil
+	}
+
+	// Also look into the path as some distributions install catatonit in
+	// /usr/bin.
+	catatonitPath, err := rtConfig.FindHelperBinary("catatonit", true)
+	if err != nil {
+		return "", fmt.Errorf("finding pause binary: %w", err)
+	}
+
+	buildContent := fmt.Sprintf(`FROM scratch
+COPY %s /catatonit
+ENTRYPOINT ["/catatonit", "-P"]`, catatonitPath)
+
+	tmpF, err := ioutil.TempFile("", "pause.containerfile")
+	if err != nil {
+		return "", err
+	}
+	if _, err := tmpF.WriteString(buildContent); err != nil {
+		return "", err
+	}
+	if err := tmpF.Close(); err != nil {
+		return "", err
+	}
+	defer os.Remove(tmpF.Name())
+
+	buildOptions := buildahDefine.BuildOptions{
+		CommonBuildOpts: &buildahDefine.CommonBuildOptions{},
+		Output:          imageName,
+		Quiet:           true,
+		IgnoreFile:      "/dev/null", // makes sure to not read a local .ignorefile (see #13529)
+		IIDFile:         "/dev/null", // prevents Buildah from writing the ID on stdout
+	}
+	if _, _, err := rt.Build(context.Background(), buildOptions, tmpF.Name()); err != nil {
+		return "", err
+	}
+
+	return imageName, nil
+}
diff --git a/pkg/specgen/generate/pod_create.go b/pkg/specgen/generate/pod_create.go
index fce32d688..5b7bb2b57 100644
--- a/pkg/specgen/generate/pod_create.go
+++ b/pkg/specgen/generate/pod_create.go
@@ -2,13 +2,8 @@ package generate
 
 import (
 	"context"
-	"fmt"
-	"io/ioutil"
 	"net"
-	"os"
 
-	buildahDefine "github.com/containers/buildah/define"
-	"github.com/containers/common/pkg/config"
 	"github.com/containers/podman/v4/libpod"
 	"github.com/containers/podman/v4/libpod/define"
 	"github.com/containers/podman/v4/pkg/domain/entities"
@@ -17,98 +12,18 @@ import (
 	"github.com/sirupsen/logrus"
 )
 
-func buildPauseImage(rt *libpod.Runtime, rtConfig *config.Config) (string, error) {
-	version, err := define.GetVersion()
-	if err != nil {
-		return "", err
-	}
-	imageName := fmt.Sprintf("localhost/podman-pause:%s-%d", version.Version, version.Built)
-
-	// First check if the image has already been built.
-	if _, _, err := rt.LibimageRuntime().LookupImage(imageName, nil); err == nil {
-		return imageName, nil
-	}
-
-	// Also look into the path as some distributions install catatonit in
-	// /usr/bin.
-	catatonitPath, err := rtConfig.FindHelperBinary("catatonit", true)
-	if err != nil {
-		return "", fmt.Errorf("finding pause binary: %w", err)
-	}
-
-	buildContent := fmt.Sprintf(`FROM scratch
-COPY %s /catatonit
-ENTRYPOINT ["/catatonit", "-P"]`, catatonitPath)
-
-	tmpF, err := ioutil.TempFile("", "pause.containerfile")
-	if err != nil {
-		return "", err
-	}
-	if _, err := tmpF.WriteString(buildContent); err != nil {
-		return "", err
-	}
-	if err := tmpF.Close(); err != nil {
-		return "", err
-	}
-	defer os.Remove(tmpF.Name())
-
-	buildOptions := buildahDefine.BuildOptions{
-		CommonBuildOpts: &buildahDefine.CommonBuildOptions{},
-		Output:          imageName,
-		Quiet:           true,
-		IgnoreFile:      "/dev/null", // makes sure to not read a local .ignorefile (see #13529)
-		IIDFile:         "/dev/null", // prevents Buildah from writing the ID on stdout
-	}
-	if _, _, err := rt.Build(context.Background(), buildOptions, tmpF.Name()); err != nil {
-		return "", err
-	}
-
-	return imageName, nil
-}
-
-func pullOrBuildInfraImage(p *entities.PodSpec, rt *libpod.Runtime) error {
-	if p.PodSpecGen.NoInfra {
-		return nil
-	}
-
-	rtConfig, err := rt.GetConfigNoCopy()
-	if err != nil {
-		return err
-	}
-
-	// NOTE: we need pull down the infra image if it was explicitly set by
-	// the user (or containers.conf) to the non-default one.
-	imageName := p.PodSpecGen.InfraImage
-	if imageName == "" {
-		imageName = rtConfig.Engine.InfraImage
-	}
-
-	if imageName != "" {
-		_, err := rt.LibimageRuntime().Pull(context.Background(), imageName, config.PullPolicyMissing, nil)
-		if err != nil {
-			return err
-		}
-	} else {
-		name, err := buildPauseImage(rt, rtConfig)
-		if err != nil {
-			return fmt.Errorf("building local pause image: %w", err)
-		}
-		imageName = name
-	}
-
-	p.PodSpecGen.InfraImage = imageName
-	p.PodSpecGen.InfraContainerSpec.RawImageName = imageName
-
-	return nil
-}
-
 func MakePod(p *entities.PodSpec, rt *libpod.Runtime) (*libpod.Pod, error) {
 	if err := p.PodSpecGen.Validate(); err != nil {
 		return nil, err
 	}
 
-	if err := pullOrBuildInfraImage(p, rt); err != nil {
-		return nil, err
+	if !p.PodSpecGen.NoInfra {
+		imageName, err := PullOrBuildInfraImage(rt, p.PodSpecGen.InfraImage)
+		if err != nil {
+			return nil, err
+		}
+		p.PodSpecGen.InfraImage = imageName
+		p.PodSpecGen.InfraContainerSpec.RawImageName = imageName
 	}
 
 	if !p.PodSpecGen.NoInfra && p.PodSpecGen.InfraContainerSpec != nil {
@@ -180,6 +95,11 @@ func createPodOptions(p *specgen.PodSpecGenerator) ([]libpod.PodCreateOption, er
 			options = append(options, libpod.WithPodUser())
 		}
 	}
+
+	if len(p.ServiceContainerID) > 0 {
+		options = append(options, libpod.WithServiceContainer(p.ServiceContainerID))
+	}
+
 	if len(p.CgroupParent) > 0 {
 		options = append(options, libpod.WithPodCgroupParent(p.CgroupParent))
 	}
diff --git a/pkg/specgen/podspecgen.go b/pkg/specgen/podspecgen.go
index 1bb64448f..603506241 100644
--- a/pkg/specgen/podspecgen.go
+++ b/pkg/specgen/podspecgen.go
@@ -204,6 +204,9 @@ type PodSpecGenerator struct {
 	PodStorageConfig
 	PodSecurityConfig
 	InfraContainerSpec *SpecGenerator `json:"-"`
+
+	// The ID of the pod's service container.
+	ServiceContainerID string `json:"serviceContainerID,omitempty"`
 }
 
 type PodResourceConfig struct {